Each time FishEye needs the contents of a revision, it fetches it from the underlying SCM system.

FishEye should use some sort of per-repository, LRU, space-limited cache of the contents of revisions.
We should independently cache the text and binary versions of files (text versions are requested much more frequently)
Consider gz compression of files.
Consider reusing the file-on-disk when returning the StreamIndexedLineReader so as to reduce double-copying.
The cache state should be preserved across FishEye restarts.

The default size of this cache (per repository) should be 500MB. It should be customizable in the admin.